Automated Kerogen Classification in Microscope Images of Dispersed Kerogen Preparation

We develop the classification part of a system that analyses transmitted light microscope images of dispersed kerogen preparation. The system automatically extracts kerogen pieces from the image and labels each piece as either inertinite or vitrinite. The image pre-processing analysis consists of background removal, identification of kerogen material, object segmentation, object extraction (individual images of pieces of kerogen) and feature calculation for each object. An expert palynologist was asked to label the objects into categories inertinite and vitrinite, which provided the ground truth for the classification experiment. Ten state-of-the-art classifiers and classifier ensembles were compared: Naïve Bayes, decision tree, nearest neighbour, the logistic classifier, multilayered perceptron (MLP), support vector machines (SVM), AdaBoost, Bagging, LogitBoost and Random Forest. The logistic classifier was singled out as the most accurate classifier, with an accuracy greater than 90. Using a 10 times 10-fold cross-validation provided within the Weka software, we found that the logistic classifier was significantly better than five classifiers (p<0.05) and indistinguishable from the other four classifiers. The initial set of 32 features was subsequently reduced to 6 features without compromising the classification accuracy. A further evaluation of the system alerted us to the possible sensitivity of the classification to the ground truth that might vary from one human expert to another. The analysis also revealed that the logistic classifier made most of the correct classifications with a high certainty.     

基于自平衡力的弹塑性动力反应分析方法

本文给出了一个基于自平衡力的弹塑性结构动力反应分析的新方法，详细叙述了方法的原理和过程，并对用改进的Ｔａｋｅｄａ滞变模型描述的梁单元框架结构给出了具体的分析方法。这个方法与传统的方法不同，在分析中不必修正结构的刚度矩阵，而代之以计算代表塑性铰变形引起的应力重分布的动力自平衡力，因而大大地节省了计算机时，最后通过算例对本文方法进行了验证。     

基于SHAP值机器学习的江西暖季暴雨预报因子重要性分析

机器学习模型（Machine Learning,ML）的不可解释性给其在气象业务中的应用带来了挑战。模型解释和可视化是解决这一问题的有效途径。文中将SHAP值应用于天气预报ML模型解释,研究了江西省暖季暴雨模型的预报因子对预报结果的影响。分别选取2016—2020年、2021—2022年4—9月ECWMF（European Centre for Medium-Range Weather Forecasts）高分辨率数值模式物理量及国家站降水观测数据进行XGBoost 建模与模型解释。结果表明,全局重要性排名前4位依次是总降水（重要性42.70%）、850 hPa比湿（重要性11.17%）、925 hPa相对湿度（重要性10.44%）、500 hPa相对湿度（重要性 9.16%）。个例分析表明,命中个例中高重要性物理因子在暴雨区的 SHAP 值较大,漏报（空报）个例在漏报（空报）区域高重要性物理因子的SHAP值偏小（偏大）。SHAP值从全局和局部可定量给出ML模型有物理意义的解释,解释结果与天气学原理和业务经验较一致,有利于ML在气象业务中的深入应用。     

一种基于核学习的储集层渗透率预测新方法

基于核学习的支持向量机,是一种采用结构风险最小化原则代替传统经验风险最小化原则的新型统计学习方法,具有完备的理论基础。这里提出了核学习技术在储集层非均质特性描述中渗透率参数预测的新用途。在复杂地层中,基于支持向量机的智能和自适应模式识别能力而建立了常规测井多参数信息输入的渗透率预测模型,然后对实际油田储集层渗透率进行了预测。与常规线性回归模型预测结果相对比,所提出的方法更易于使用,很少受不确定因素的影响,并具有较强的信息整合能力以及更高的预测准确性和可信度。     

基于Faster R-CNN的食品图像检索和分类

面向食品领域的图像检索和分类等方面的研究成为多媒体分析和应用领域越来越受关注的研究课题之一.当前的主要研究方法基于全图提取视觉特征,但由于食品图像背景噪音的存在使得提取的视觉特征不够鲁棒,进而影响食品图像检索和分类的性能.为此,本文提出了一种基于Faster R-CNN网络的食品图像检索和分类方法.首先通过Faster R-CNN检测图像中的候选食品区域,然后通过卷积神经网络（CNN）方法提取候选区域的视觉特征,避免了噪音的干扰使得提取的视觉特征更具有判别力.此外,选取来自视觉基因库中标注好的食品图像集微调Faster R-CNN网络,以保证Faster R-CNN食品区域检测的准确度.在包括233类菜品和49 168张食品图像的Dish-233数据集上进行实验.全面的实验评估表明：基于Faster R-CNN食品区域检测的视觉特征提取方法可以有效地提高食品图像检索和分类的性能.     

自调节趋势分析法

自调节趋势分析法的特点是:能自动消除局部成分对计算拟合区域成分的多项式的系数的影响,并能确定多项式的阶次,因而在地学数据处理及预测中有广泛的应用。     

基于迁移学习的地球物理测井储层参数预测方法研究

随着大数据和机器学习的成熟和推广应用,人工神经网络在地球物理测井预测储层参数中得到重视.本文引入迁移学习进行测井储层参数预测,以孔隙度预测神经网络模型和孔隙度含水饱和度联合预测神经网络模型为基础模型,分别以渗透率及含水饱和度预测作为目标任务进行迁移学习,以提升储层参数预测效果和效率.文中详细阐述了基于迁移学习的测井储层参数预测方法,并使用64口井的测井数据进行储层参数预测效果分析.结果表明,使用迁移学习后,渗透率模型预测效果最高可以提升58.3%;含水饱和度模型预测效果最高可以提升近40%,且最大可以节省60%的计算资源;以孔隙度预测模型为基础模型时更适合使用参数冻结的训练方式,以孔隙度含水饱和度联合预测模型为基础模型时更适合使用参数微调的训练方式.

     

基于半监督学习的井震联合储层横向孔隙度预测方法

传统地震储层预测技术一般基于弹性参数反演和岩石物理建模的级联流程实现储层孔隙度预测, 其预测精度受到波动理论和岩石物理理论的近似假设、初始模型和二次反演累积误差等因素的影响.为缓解这些问题, 本文提出了一种基于双向门控递归单元神经网络的半监督学习井震联合孔隙度预测方法, 实现从地震数据直接预测储层横向孔隙度.通过少量的地震测井样本标签对和多目标函数约束建立智能化多尺度多信息融合孔隙度预测模型, 实现地震数据到孔隙度, 孔隙度再到生成地震数据的闭环映射.此外, 在网络模型每次迭代更新的过程中随机引入非井旁地震道参与网络训练, 非井旁地震道的波形匹配能在一定程度上保证井间孔隙度的预测精度.模型数据和实际数据测试结果表明, 本文提出的方法相比于有监督学习孔隙度预测方法能进一步提高储层孔隙度的预测准确性和横向连续性, 获得较为可靠的储层物性参数的空间分布.

     

The use of digital technologies for landslide disaster risk research and disaster risk management: progress and prospects

Landslide is a serious natural disaster next only to earthquake and flood, which will cause a great threat to people’s lives and property safety. The traditional research of landslide disaster based on experience-driven or statistical model and its assessment results are subjective , difficult to quantify, and no pertinence. As a new research method for landslide susceptibility assessment, machine learning can greatly improve the landslide susceptibility model’s accuracy by constructing statistical models. Taking Western Henan for example, the study selected 16 landslide influencing factors such as topography, geological environment, hydrological conditions, and human activities, and 11 landslide factors with the most significant influence on the landslide were selected by the recursive feature elimination (RFE) method. Five machine learning methods [Support Vector Machines (SVM), Logistic Regression (LR), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Linear Discriminant Analysis (LDA)] were used to construct the spatial distribution model of landslide susceptibility. The models were evaluated by the receiver operating characteristic curve and statistical index. After analysis and comparison, the XGBoost model (AUC 0.8759) performed the best and was suitable for dealing with regression problems. The model had a high adaptability to landslide data. According to the landslide susceptibility map of the five models, the overall distribution can be observed. The extremely high and high susceptibility areas are distributed in the Funiu Mountain range in the southwest, the Xiaoshan Mountain range in the west, and the Yellow River Basin in the north. These areas have large terrain fluctuations, complicated geological structural environments and frequent human engineering activities. The extremely high and highly prone areas were 12043.3 km² and 3087.45 km², accounting for 47.61% and 12.20% of the total area of the study area, respectively. Our study reflects the distribution of landslide susceptibility in western Henan Province, which provides a scientific basis for regional disaster warning, prediction, and resource protection. The study has important practical significance for subsequent landslide disaster management.     

融合先验物理信息的高精度智能可控源电磁反演算法

可控源电磁反演利用人工信号获取地下结构信息,为地质勘探和资源开发提供准确的数据支持。然而,传统电磁反演方法面临低分辨率的挑战,主要是由于简化处理和观测数据的限制,导致模型平滑、细节丢失,从而削弱了反演的准确性,影响了电磁勘探的效率。为了解决这一问题,本文提出将传统反演结果与响应数据同时作为深度网络的输入数据,为深度网络反演提供先验物理信息,并结合深度学习算法提升可控源电磁反演的计算效率。通过模型试验,对合成的电阻率模型分别进行了传统反演、智能反演以及融合先验物理信息的智能反演。结果表明,融合先验物理信息的智能反演能够更好地刻画异常体结构特征,有效提升反演效率,并且得到的电阻率参数更符合实际。最后,将该反演技术应用于金川铜镍矿床的可控源数据反演解释,取得了较为可靠的应用效果。